PRODUCT | 产品中心 |
CSD 2000 CSD 概观 CSS 1000 CSS 概观 APPLICATIONS 应用价值
Terabytes of 3D NAND - ScaleFlux Computational Storage
Terabytes of
3D Flash
高密度闪存
3D Flash
Compute Enginer Acceleration - ScaleFlux Computational Storage
Compute Engine
Acceleration
计算引擎
加速
ScaleFlux Computational Storage Cards - PCIe AIC & U.2 Drive
ScaleFlux Computational Storage ScaleFlux 计算存储
PCIe HHHL AIC and U.2 PCIe HHHL AIC 和 U.2

The Need for Computational Storage 计算存储

Optimize the Scaling Trajectory 优化数据基础设施的拓展轨迹

Modern enterprise and cloud infrastructure is faced with the daunting task of scaling compute and storage infrastructure to respond to the variety, velocity, and volume growth of data. With the demise of Moore’s Law, adding millions of storage IOPS to servers via SSDs has exposed processing bottlenecks that limit overall server and storage scaling. CSS integrates value-add compute engines with high capacity Flash storage, dramatically increasing the processing efficiency of compute-intensive functions. 当今企业和云计算数据中心正面临着迅猛数据量增长和随之带来数据存储和计算负荷的重大挑战。对数据体量,处理速度和多样性的应变能力,是企业和云计算数据中心核心竞争力的体现。随着摩尔定律的衰减,SSD为服务器增加的百万计的IOPS,加之以太网带宽的迅速增长,使得数据计算和处理逐渐成为瓶颈,从而限制的服务器和存储的进一步提升。 CSS 融合了高性能计算引擎和高密度闪存, 显著提升重负荷计算的效率。

#compute2data #compute2data 创新的数据流动架构

CSS brings heavy-duty compute engines right to the data and uniquely parallelizes compute resources across multiple Flash-based storage devices connected through PCIe. Compute resources scale as additional Flash storage is added. This alleviates CPU processing limitations and helps to reduce the highly expensive task of moving Petabytes of data from storage to the processor for computation. CSS is the foundation for modern, data- driven Enterprise and Cloud infrastructure. CSS 让重负荷计算更加靠近数据,通过PCIe端的连接,在多个闪存存储设备上得以并行增加计算资源。其效果是额外闪存的增加亦带来计算资源的增加。这极大地减轻了服务器CPU的负载,也大幅度减少了数据因计算带来的的大量搬移。CSS将为企业和云计算提供领先的基石。

TCA/TCO Savings and Agility for an Evolving Data Center 节省TCA/TCO成本,灵活适配数据中心增长

CSS integrates state-of-the-art Flash controllers with LDPC error correction enabling the latest and most cost-effective 3D NAND Flash to be deployed in Enterprise environments. CSS enables field programmable and updateable compute engines through a soft-hardware platform approach. This allows the infrastructure to evolve in lock step with new application demands. CSS包含了业界最为领先的自研闪存控制器,具备超强的LDPC纠错和信号处理功能,使得数据中心以最快的速度部署最新的低成本3D NAND闪存成为可能。CSS同时使得高性能计算引擎可在服务器内实时定制和更新,从而让数据基础设施能对新的业务和应用及时适配所需的计算功能。

Revolutionary Performance, Evolutionary Integration 简化集成,提升性能

CSS is integrated into x86/Linux server and storage environments via an easy-to-install ScaleFlux software module. Host-based Flash Translation Layer and Flash Management provides consistent latency and performance. CSS Compute Engines are accessible to applications through familiar APIs exposed from the software module. By simultaneously solving compute and storage I/O bottlenecks, CSS provides significant and proven run-time improvements to compute and data intensive applications. 通过简单易用的软件安装模块,CSS可以无缝的集成到 x86/Linux 通用服务器和存储环境。服务器端的闪存映射和管理为低延时,高性能和高稳定性的应用提供了保障。用户可通过API灵活调用CSS的高性能计算引擎. 通过同时解决计算和存储的I/O瓶颈,CSS为大数据的处理和应用的大幅度性能提升和成本降低提供了坚定的基石。

New 2.3.X Features New 2.3.X Features

Available to new and existing customers in software release 2.3.X! Available to new and existing customers in software release 2.3.X!

We are proud to introduce Atomic Write and Namespace Streams to the CSS 1000 Series. Thanks to the reprogrammable nature of CSS' FPGA-based platform, these updates will roll out to all existing customers via a simple downloadable upgrade. We are proud to introduce Atomic Write and Namespace Streams to the CSS 1000 Series. Thanks to the reprogrammable nature of CSS' FPGA-based platform, these updates will roll out to all existing customers via a simple downloadable upgrade.

Atomic Write Atomic Write

Atomic write (AW) is the ability to guarantee a single write of blocks up to 1MB in size. For database applications, this assures each Database record update is written in its entirety in a single transaction rather than breaking up the write into multiple transactions. This bypasses the traditionally used Double Write Buffer and increases drive endurance as a result of reducing total write operations. AW preserves existing benefits such as recovering data in the event of a crash and increases transactions per second for write-intensive workloads. Atomic write (AW) is the ability to guarantee a single write of blocks up to 1MB in size. For database applications, this assures each Database record update is written in its entirety in a single transaction rather than breaking up the write into multiple transactions. This bypasses the traditionally used Double Write Buffer and increases drive endurance as a result of reducing total write operations. AW preserves existing benefits such as recovering data in the event of a crash and increases transactions per second for write-intensive workloads.

Namespace Streams Namespace Streams

Namespace Streams (NS) greatly improves Flash storage I/O performance, latency and endurance by minimizing unnecessary garbage collection. NS allows customers to create separate Streams of logically-associated data with different life cycles by simply associating each Stream with a Namespace. For example, log updates, files, and transaction records could each be a Stream. CSS then places the data from a given Stream together on physically-associated region of Flash, each with its own set of garbage collection units and its own tunable overprovisioning. This yields lower overall latency and write amplification. Namespace Streams (NS) greatly improves Flash storage I/O performance, latency and endurance by minimizing unnecessary garbage collection. NS allows customers to create separate Streams of logically-associated data with different life cycles by simply associating each Stream with a Namespace. For example, log updates, files, and transaction records could each be a Stream. CSS then places the data from a given Stream together on physically-associated region of Flash, each with its own set of garbage collection units and its own tunable overprovisioning. This yields lower overall latency and write amplification.

Key Features 基本特征

  • Form Factor
    PCIe AIC & U.2 Drive
  • Flash Capacity
    1.6, 1.92, 3.2, 3.84, 6.4, 7.68TB
  • Interface
    PCIe Gen3 x4
    Low-latency flash storage I/O device
    APIs/Libraries for Compute Acceleration integration
  • Compute Engines
    GZIP Compression, Erasure Coding, Flexible KV-Store Interface, Customizable Database Engine Accelerator
  • Tunable FTL/FM
    Adjust drive parameters to fine-tune performance
  • Atomic Write
    Guarantee a single, complete write operation for up to 1MB I/Os to avoid double-write buffer penalties to performance & endurance
  • Namespace Streams
    Separate resource pools for each Stream of data to eliminate unnecessary Garbage Collection and improve Quality of Service
  • Temp & Power Throttle
    End-to-end data protection and ECC (Error Correction Code) on all internal memories in the data path; Integrated LDPC engine and Flash die RAID assures 10-20 UBER
  • Power Loss Protection
    Complete data protection from unplanned power loss
  • In-Place Updates
    Evolve your infrastructure with new applications through in-system - programmability of CSS hardware, just like with software

Technical Specifications 产品规范

Please contact us for more info. 联系我们

Partnerships 伙伴关系

Worldwide Availability

ScaleFlux Computational Storage Drives can be conveniently ordered from our Value Added Reseller or Distributor partners and are qualified on an ever-growing list of system OEMs.


CSD in the Cloud

ScaleFlux has also partnered with Packet to bring the acceleration of Computational Storage to the convenient public cloud.


Additional Resources 额外的文件

ScaleFlux Product Brief (EN) ScaleFlux Product Brief (EN)
Information about our Computational Storage Subsystem 1000 Series product line. Information about our Computational Storage Subsystem 1000 Series product line.